"Ink Core Artificial Intelligence" completed hundreds of millions of A rounds of financing and developed the first cloud AI chip with sparse architecture

Recently, 36Kr learned that Moxin Artificial Intelligence Technology (Shenzhen) Co., Ltd. (hereinafter referred to as "Moxin Artificial Intelligence") has obtained hundreds of millions of yuan A round of financing, which is led by Cornerstone Capital and the Greater Bay Area Common Home Development Fund, and co-invested by Tongwei Capital, Zhongke Huachuang Investment Fund and Shenzhen Angel Fund of Funds.

It is reported that this round of financing will be mainly used for the mass production of the first chip and the expansion of the sparse ecosystem. China Renaissance Capital serves as the exclusive FA for this round.

Moxin Artificial Intelligence Technology was founded in Silicon Valley in 2018 and is currently headquartered in Shenzhen. Ink core through the sparse algorithm to create high-performance low-cost AI computing power, its products are currently mainly accelerator chips, mainly used in data centers.

According to IDC data, the global AI chip market will reach $35.2 billion by 2022, with a compound annual growth rate of more than 55%. At present, the Ai chip market in China and even the world is still in its early stages, and there are still great opportunities.

Wang Wei, founder and CEO of Moxin Artificial Intelligence, told 36Kr that the current deep learning algorithm achieves high precision on a single task, and considering that the future will be a multi-task giant neural network, sparse neural network will be the development trend of deep learning. Since the chips on the market cannot support the sparse matrix algorithm very well, the ink core chooses to independently develop a computing chip that can provide the underlying technology for it.

The ink core artificial intelligence team believes that sparseness is the development direction of AI chips in the future.

Ink core artificial intelligence has dynamic sparse and static sparse ("double sparse") technology, starting from algorithms and software, defining hardware by algorithm software, and taking the soft and hard collaborative route.

The company developed a chip Inteng processor ANTOUM, which supports up to 32X sparse rates. At present, the chip has been returned from the TSMC current chip, and the first power-on on the day was successfully lit within a few seconds, and in less than 48 hours, the basic test of all functional modules was completed synchronously and the full model of ResNet50 was run through.

Wang Wei introduced that the energy efficiency ratio of the Antoum processor (ANTOUM) is 1 order of magnitude higher than that of the mainstream products in the industry, and the single card computing power is increased by 5-10 times. In addition, it supports the mainstream development framework and a wide range of AI operator libraries and models in the current market, and can compile more than 200 neural network models, that is, ANTOUM chips provide high computing power and high energy efficiency while also having versatility at the software level.

Based on the chip, Moxin Artificial Intelligence has developed two AI computing acceleration cards, the main low-power and cost-effective SparseOne (half-height and half-length), and the high-computing Sparse Megatron (full height and full length), which will be released in the first quarter of 2022.

In the early days, the company focused on data center scenarios, providing services for three types of users, such as the Internet, operators (such as telecom, mobile, Unicom, etc.) and government/security private cloud.

Wang Wei told 36Kr that due to the company's innovation in the sparse algorithm level, the company can communicate and cooperate with the head customers in the fields of the Internet, operators and government security in the early stage, which can enable the ink core to define the product more accurately according to the needs of users, and also deepen the understanding of potential customers on the ink core. It is worth noting that Inspur and Smart Interconnection Industry Fund are early strategic investors in Ink Core.

In the founding team, the company's founder and CEO Wang Wei is a Master of ECE from Carnegie Mellon University, a Silicon Valley chip expert with nearly 20 years of experience, a former Qualcomm and Intel core architect, and one of the core teams of Intel's fifth- to tenth-generation CPU processors, participating in the cumulative mass production of more than 5 billion chips. He also has several years of entrepreneurial experience with Silicon Valley chip companies. Co-founder and chief scientist Eun-hyun Yan, a Ph.D. in machine learning at Carnegie Mellon University, has worked as a researcher at Google and Microsoft Research, and between 2015 and 2019, he was the first author to publish the second most articles in the world at NIPS.

In terms of team composition, the ink core artificial intelligence team is composed of two parts, one is from carnegie Mellon University top AI scientists, the other from the world's top semiconductor companies (such as Intel, Qualcomm, Marvell and Oracle, etc.) core high-volume chip research and development team, the company's hardware, software, architecture, marketing, sales and other fields of managers have experience in the top mass production chip industry experience.

At present, Moxin is headquartered in Shenzhen, with R&D branches in Shanghai, Beijing and Silicon Valley, and the company has approached the scale of 100 people.